1. Technical Field
The present invention relates to object detection apparatuses that detect objects based on image data.
2. Description of the Related Art
Conventionally, there has been proposed a technique of: scanning image data through windows of a predetermined size; calculating for each window a first evaluation value which represents the probability of an object being included in the window; extracting candidate regions based on the first evaluation values calculated for the windows; calculating for each candidate region a second evaluation value which represents the probability of the object being included in the candidate region; and extracting, based on the second evaluation values calculated for the candidate regions, an image region in which the object is included (see, for example, Japanese Patent Application Publication No. JP2009059047A).
That is, the conventional technique is a technique of performing a process of detecting an object by serially linking (or connecting) plural means for detecting the object. In this case, each detecting means extracts a desired image region by comparing the first evaluation values or the second evaluation values with a preset threshold.
To realize a thorough detection, it is preferable to set the threshold for the first-stage (or earlier-stage) detecting means to a low value. However, with the low threshold, the number of the candidate regions extracted by the first-stage detecting means, i.e., the number of the candidate regions to be processed by the second-stage (or later-stage) detecting means is increased, thereby increasing the processing time. In contrast, if the threshold for the first-stage detecting means is set to a high value for the purpose of reducing the processing time, it may be difficult to realize a thorough detection. That is, there exists a trade-off between reduction in the processing time and the realization of a thorough detection.
Moreover, the number of the candidate regions extracted by the first-stage detecting means varies greatly depending on the image data inputted to the first-stage detecting means.
Furthermore, in the case of, for example, an in-vehicle system, it is necessary to obtain the processing results in real-time and thus there is a limitation on the processing time; and the processing capacity of a processor provided in the system for performing the processing task is limited. In such a case, to complete the processing task within a time limit regardless of the size of the image data, it is necessary to set the threshold for the first-stage detecting means to a high value. However, with the high threshold, there is generally caused leeway in the processing time, and the leeway is generally not effectively utilized.